Kimi K3
- Made by
- Moonshot AI
- Released
- Announced in
- Kimi K3: Open Frontier Intelligencekimi.ai
- Event
- Kimi K3
- API record
- /api/v1/models/kimi-k3
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| Humanity's Last ExamNo tools · full set; max reasoning effort | 43.5% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| Humanity's Last ExamWith tools · full set; max reasoning effort | 56.0% | CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| GPQA Diamondmax reasoning effort | 93.5% | CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| BrowseCompmax reasoning effort; context compaction at 300K tokens | 91.2% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| 1M-token context, no context management | 90.4% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| AutomationBench600-task public subset; max reasoning effort | 30.8% | CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| DeepSWE v1.1Kimi Code harness, max effort | 67.5% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| mini-SWE-agent harness, as on the official leaderboard | 67.3% | CheckedA person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| MCP Atlasmax reasoning effort; 500-task public set; 100-turn limit; Gemini 3.1 Pro judge | 84.2% | CheckedA person checked this number against its source on 18 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
Sources
Reported by Moonshot AI
- Kimi K3: Open Frontier Intelligencekimi.ai · 16 Jul 2026
Our capture of the source